Lecture 15

Power system state estimation

Kun Zhu
2013-05-02
 Outline

State estimation
- What
- Why

Weighted Leaset Square (WLS) algorithms

- Mathmatics
- Concepts

Operation challenges
Course road map
What is state estimation?
How could the operator know the
system?

62 MW

6 MW
 The truth is out here!

Courtesy of ABB Ventryx

Power system states: x

The power system states are those parameters that can be

used to determine all other parameters of the power system

Node voltage phasor

- Voltage magnitude Vk
- Phase angle k
Transformer turn ratios
- Turn ratio magnitude tkn
- Phase shift angle kn
Complex power flow
- Active power flow Pkn, Pnk
- Reactive power flow Qkn, Qnk
Analog measurements

Voltage magnitude
Current flow magnitude & injection
Active & reactive power
- Branches & groups of branches
- Injection at buses
- In switches
- In zero impedance branches
- In branches of unknown impedance
Transformers
- Magnitude of turns ratio
- Phase shift angle of transformer
Synchronized phasors from
Phasor Measurement Unit
Network topology processing

Bus breaker model Bus branch model

 Power system measurements: z

z=ztrue+e

z=ztrue+e
ztrue: power system truth
ztrue=h(x)

e: measurement error

e=esystematic+erandom z=ztrue+e
Measurement model

How to determine the states (x) given a set of

measurements (z)?
zj = hj(x) + ej
known unknown unknown

where
- x is the true state vector [V1,V2,Vk, 1, 2, k]

- zj is the jth measurement

-hj relates the jth measurement to states
-ej is the measurement error
State estimation process
Why do we need state estimation?
Measurements correctness

Imperfections in
- Current & Voltage transformer
- Transducers
A/D conversions
Tuning
- RTU/IED Data storage
- Rounding in calculations
- Communication links

Result in uncertainties in the measurements

Measurement timeliness

Due to imperfections in SCADA system the

measurements will be collected at different points
in time, time skew.

Pab
Pbc
Pcb
Pdc
t

If several measurements are missing how long to
wait for them?
Fortunately, not a problem during quansi-steady
state.
State estimation is used for off-line applications
How can the states be estimated?
Approaches
Minimum variance method
- Minimize the sum of the squares of the weighted
deviations of the state calculated based on
measurements from the true state
Maximum likelihood method
- Maximizing the probability that the estimate equals
to the true state vector x
Weighted least square method (WLS)
- Minimize the sum of the weighted squares of the
estimated measurements from the true state

m
( zi hi ( x))2
J ( x) =
i =1 Rii
Least square (Wiki)

"Least squares" means that the overall solution

minimizes the sum of the squares of the errors made
in the results of every single equation.
The method of least squares is a standard approach
to the approximate solution of over determined
system, i.e., sets of equations in which there are more
equations than unknowns.
The most important application is in data fitting.
Carl Friedrich Gauss is credited with developing the
fundamentals of the basis for least-squares analysis in
1795.
WLS state estimation

Fred Schweppe introduced state estimation to power

systems in 1968.
He defined the state estimator as a data processing
algorithm for converting redundant meter readings
and other available information into an estimate of the
state of an electric power system.
Today, state estimation is an essential part in almost
every energy management system throughout the
world.

Felix F. Wu, Power system state estimation: a survey,

International Journal of Electrical Power & Energy
Systems, Volume 12, Issue 2, April 1990, Pages 8
WLS state estimation model

zj = hj(x) + ej
known unknown unknown
Error characteristics

The errors in the measurements is the sum of

several stochastic variables
- CT/VT, Transducer, RTU, Communication
The errors is assumed as a Gaussian Distribution
with known deviations
- Expected valueE[ej] =0
- Known deviation j
The errors are also assumed to be independent
- E [eiej] =0
 WLS objective function

m 2
( zi h( x)) T 1
J ( x) = 2
= [ z h( x)] R [ z h( x)]
i =1 i

where
i=1,2,m

R = diag{ 12 , 22 , 32 ,... m2 } = Cov(e) = E[e eT ]

Solution to above is iterative using newton methods
 Newton iteration

At the minimum, the first-order optimality conditions

will have to be satisfied

J ( x) T 1
g ( x) = = H ( x) R [ z h( x)] = 0
x
h( x)
H T ( x) = [ ] is the measurement Jacobian matrix
x
Newton iteration contd

Expanding the g(x) into its Taylor series around state

k
vector x

k k k
g ( x) = g ( x ) + G( x )( x x ) + ..... = 0
where

k
k g ( x ) T k 1 k
G( x ) = = H ( x )R H ( x )
x
Newton iteration contd

Neglecting the higher order terms leads to an iterative

solutions scheme know as the Gauss-Newton method
as :

k +1 k k 1 k
x = x [G( x )] g ( x )
k is the iteration index,
xk is the solution vector at iteration k






Newton iteration IV

Convergence

k
max( x )

If not, update
k +1 k k
x = x + x
k = k +1
Go back to the previous step

Why the measurements are
weighted?

voltage mag

complex power flow

Pij = f (vi , v j , i , j )
Qij = f (vi , v j ,i , j )
Weight

Weight is introduced to emphasize the trusted

measurement while de-emphasize the less
trusted ones.

WLS
1
Wi = 2
i
Observability

Based on system topology

and location of
measurements parts of the
power system may be
unobservable.

Unobservable parts of the

system can be made
observable via data
exchange (CIM), pseudo
measurements, etc
Observability example

voltage mag

complex power flow

Pij = f (vi , v j , i , j )
Qij = f (vi , v j , i , j )
Observability criterion

Necessary but not sufficient condition

mn
m: Number of measurements
n: Number of states

Is the systems oservability guranteed in this

case?
Observability example II

voltage mag

complex power flow

Pij = f (vi , v j , i , j )
Qij = f (vi , v j ,i , j )
n of m measurements has to be independent

voltage mag

complex power flow

Pij = f (vi , v j , i , j )
Qij = f (vi , v j ,i , j )
Summary of assumptions

Quansi-steady system
- No large variations of states over time
- State estimator will be suspended when large
disturbance happens.

Errors in measurements are

- Gaussian in nature with known deviation
- Independent

Strong assumption
- Power system topology model is correct
Refinements

Refinements of method for State Estimation has

been the objective of much research.

Reduce the numerical calculation complexity in

order to speed up the execution.
- 3000 bus node in 1-2 seconds

Improve estimation robustness

- less affected by erroneous input
Bad Data Detection
Data quality

Analog measurement error

Parameter error
Topological error
- Discrete measurement error
- Model error
Bad Data Detection (Analog)

Putting the measurements up to a set of

logical test (Kirchhoff's laws) before they are
input into the State Estimator
Calculating J(x) and comparing with a pre-
determined limit. If the value exceeds the limit,
we can assume there is something wrong in the
measurements (Chi-square)
Bad Data Detection (Analog)

Once the system state has been identified, i.e. we

have an estimate of x

We can use the estimate to calculate

ri = z i hi (x )
If there is any residual in this calculation that
stands-out this is an indication that particular
measurement is incorrect
Largest normalized residual

N zi h( x )
ri =
i
The normalized residual follows the standard
normal distribution
Limitations

The critical measurements have zero residuals,

hereby they can not be detected

t t t t t
0 0 0 0 0
T T 1
zi = H i ( H H ) H R z = Kz S = . . . 0 .

. t . . .
r = z z = (1 K ) z = Sz
0 0 0 0 t
Critical measurement example

voltage mag

complex power flow

Pij = f (vi , v j , i , j )
Qij = f (vi , v j ,i , j )
Parameter and structural processing

zi = hi ( x) + ei zi = f i ( p) + ei

Similar theory can be applied

State estimation functions

1. Bad data processing and elimination given

redundant measurements

2. Topology processing:
create bus/branch model (similar to Y matrix)

3. Observebility analysis:
all the states in the observable islands have
unique solutions

4. Parameter and structural processing

Challenges

Perception of the process is from the

measurements whose quality is, to large
extent, out of our control

The quality of estimates relies on the input

whose uncertainty is highly dependent on the
ICT infrastructure

Power system model can contain errors